Stuart Barber's research page


List of publications


Wavelet methods in Statistics

Wavelets are a class of functions which can be used to generate orthogonal bases for spaces of functions. Because of the way wavelets are designed, the description of a "nice" function in terms of a wavelet basis tends to be very efficient. Here, "efficient" means that you only need a few wavelets to describe quite complicated functions.

Wavelets have been used in engineering, signal processing, and numerical analysis for some time. In the last fifteen years the statistical community has been applying wavelets to problems including nonparametric regression, density estimation, time series analysis, and changepoint detection.

A good website for an introduction to wavelets is Amara's wavelet page. Another nice document is Wavelets: Seeing the forest - and the trees, an article written as part of the National Academy of Sciences' "Beyond Discovery" series of articles on important scientific advances.

There are many web sites devoted to wavelets and their use, including WaveThresh (help pages), a free add-on package for wavelet analysis in R or S-Plus . Also useful are the wavelet digest and a list of wavelet researchers.

One of the interesting things about working with wavelets is the fact that they are used by researchers in many different disciplines. Two useful introductions to wavelets written from an engineering point of view are Clemens Valens' "A Really Friendly Guide to Wavelets" and Robi Polikar's The Engineer's Ultimate Guide to Wavelet Analysis.

WaveBand

There have been many wavelet thresholding rules proposed in the literature, but relatively little work has been done so far on producing confidence intervals to go with the curve estimates. One Bayesian approach is the WaveBand method described in Barber, Nason & Silverman (2002). This method has been implemented in WaveThresh; code and help pages are available on the WaveBand web page.

Here are some slides for a talk on the WaveBand method.

CThresh

The best-known wavelets in the statistical literature are the "extremal phase" and "least asymmetric" families of wavelets due to Ingrid Daubechies. It is not so well known that there are also complex-valued Daubechies wavelets (cDws). It turns out that these complex-valued wavelets are very good at denoising real-valued signals. Barber & Nason (2004) describe two denoising techniques that use these cDws. Code to use these methods in WaveThresh is nearly ready for release.

Sequential Analysis

Sequential methods are of great potential use in clinical trials. The basic idea is very simple and is based on the fact that during a clinical trial, data is steadily accumulated. Rather than waiting until the end of the trial to analyse all the data, periodic examination of the data accumulated to date means that if there is a clear difference in favour of one of the treatments the trial can be stopped earlier than planned. By allowing the possibility of stopping the trial at an early stage if there is sufficient evidence in favour of one of the treatments on trial, significant saving in time and resources can be made. Moreover, it is possible to greatly reduce the expected number of patients involved in the trial and thus minimise the number of patients who will be given an inferior treatment.

Software for sequential clinical trials

One of reason that sequential methods have not seen wider use in clinical trials is the lack of software designed to implement these methods. Commercial programs available include S+SeqTrial (Insightful) and EaSt (Cytel Software Corporation). However, some authors have also made programs available over the web. These include

There is also my own code for designing optimal symmetric and asymmetric group sequential boundaries using the approach described in Barber & Jennison (2002).


List of publications


Collaborators

Robert Aykroyd; Chris Fallaize; Mark Gilthorpe; Alex Goodwin; Richard Jackson; Christopher Jennison; John Kent; Kanti Mardia; Guy Nason; Sam Peck; Bernard Silverman; Rebecca Walls.
[University of Leeds] [School of Mathematics] [Department of Statistics] [Home page]
Last modified: Wed Jul 30 16:44:36 BST 2008